What is the Prevalence of Low Health Literacy in European Union Member States? A Systematic Review and Meta-analysis

Three reviewers searched the bibliographic databases PubMed, Embase, and Scopus using the following string: (“health literacy”[Title/Abstract]) AND (((evaluat*[Title/Abstract]) OR measure*[Title/Abstract]) OR assess*[Title/Abstract]). The string was adapted to fit the search criteria of each database (Supplementary Table 1). No reference librarian was involved. The worldwide discussion on a comprehensive HL definition started at the beginning of the twenty-first century¹⁸; therefore, all articles published between 1 January 2000 and 23 June 2019 were retrieved, without restrictions of language or paper type. The search was supplemented by scanning the reference lists of the relevant articles.

Duplicate articles were removed, and the title and abstract of all retrieved records were screened. Studies that did not meet the inclusion criteria were excluded. Full texts of potentially relevant articles were examined by three researchers and reasons for exclusion were recorded. As suggested by Jackson et al.¹⁹, we used Google Translate to assess for inclusion of non-English and non-Italian-language articles.

We included any article with the following characteristics: (i) cross-sectional design; (ii) conducted in one or more European Union countries (EU-28); (iii) quantified the prevalence of low HL using a valid and reliable tool; (iv) included people aged 18 years or over.

We excluded articles that investigated only specific HL (e.g., oral HL) that assessed only specific HL domains without providing a general measurement or that did not report the prevalence of low HL in its target population(s).

For each record, three reviewers independently extracted the following information: first author, year of publication, tool used to quantify the prevalence of people with low HL, assessment method, number of items of the tool, proportion of people with low HL, sample size, country, EU geographic area, target population, mean or median age of the sample. Supplementary Table 2 illustrates the characteristics of the HL instruments that were used to assess HL and the cut-off scores that were considered to identify people with low HL. The assessment method was classified according to the structure of the tool in four different categories: using word recognition items, using reading or numeracy comprehension items, using self-reported comprehension items, or using a mixed method (i.e., combination of self-reported and reading or numeracy comprehension items). Countries were grouped as North, East, West, and South Europe, according to the United Nations classification²⁰. The target population was classified as general population (i.e., without specific characteristics reported), oncology patients, chronic disease patients, or refugees.

Three independent authors performed the quality assessment of the articles included in the systematic review using the Newcastle-Ottawa Scale for evaluating cross-sectional/survey studies²¹. Articles were considered of high quality when the total score was ≥ 7, fair quality if the score was ≥ 5 and < 7, and poor quality if the score was lower than 5²².

Since most articles provided two or more prevalence estimates (e.g., in different populations, in different age groups, in different countries, using different tools), we considered each estimate to be a different estimate. They will hereafter be collectively referred to as “studies.”

As in a few reports the same target population was investigated using more than one HL tool, separate meta-analyses were conducted. Specifically, we performed an inverse-variance weighted meta-analysis using a logit transformation of the proportions for each HL assessment method. We pooled estimates using a random effects approach²³ and the restricted maximum likelihood method²⁴. The I² metric was used to test for heterogeneity²⁵.

A random effects meta-regression analysis using logit-transformed prevalence was performed to explore the association between study characteristics and pooled low HL prevalence^25,26. The robust variance estimation was used to take into account the correlation between studies²⁷. We ran univariate and multivariable analyses including the covariates that could influence the prevalence estimate based on literature review. The final model consisted of the following variables: geographical area, study quality, assessment method, target population, and mean/median of the sample. The category with the highest number of studies was used as reference for geographic region, target population, and study quality; for the assessment method, we used the category yielding the lowest illiteracy pooled estimate; for the age groups, we followed the natural gradient, using the youngest as reference. For the breakdown of the age categories, we used the cut-off values reported in most studies. All analyses were performed using STATA (StataCorp), version 16.0.

After removal of duplicates, 9120 records resulted from the systematic search (Fig. 1). Screening by title and abstract yielded 134 articles that were then assessed for eligibility. A total of 62 articles were ultimately included in this systematic review, which provided the data of 101 studies (Supplementary Table 3). Since in two reports^28,29 the same target population was investigated by different tools but applying the same HL assessment method, only the prevalence estimate coming from the most frequently used tool was included in the meta-analysis, for a total of 99 studies that were pooled.

A similar number of studies was available from countries of the North^10,30‐53, South^10,54‐71, and West EU region^{10,28,72‐85} (Table 1). Conversely, only four studies were conducted in the East^10,86,87, and two studies referred to refugees from non-EU countries^29,88. A consistent heterogeneity was observed among the tools used to assess HL: the most frequently used tools were the Newest Vital Sign, applied in 21 studies^{42,43,45,47,52,55‐58,63,67,73,77,79,82,84}, followed by the European Health Literacy Survey Questionnaire with 16 items (n = 15)^{26,27,33,34,38,50,60,71,75,78,83,86} and the Rapid Estimate of Adult Literacy in Medicine (n = 12)^{30,31,40‐42,47‐49,73,76}. The European Health Literacy Survey Questionnaire with 47 items^{38,43,55,62,63,78,84,87} and its longer version with 86 items^10,58 were used eleven and nine times each, respectively, followed by the Single Item Literacy Screener (n = 8)^{44,61,65,66,69,82} and the Medical Term Recognition Test (n = 7)^56,64,65 (Table 1). The other tools were used in a limited number of studies, from one to three each (Table 1). Most studies investigated HL using self-reported comprehension items^{26,27,30,33,34,36,38,44,48‐50,54,55,58,60,61,67,68,70‐72,74‐76,78,81,83,85,86} (n = 38); 29 studies used reading or numeracy comprehension items^{35,37,39‐43,45‐47,51,52,55‐58,63,67,73,77,79,80,82,84}; 23 studies used word recognition items^{28‐32,58,59,62,65,66,79,82}; and nine studies used a mixed method^10,58. Sixty-six studies quantified low HL in the general population^{10,26,29,31‐34,36,39,40,43‐45,47,48,50‐65,70,71,73,75,76,78,80,83,84}. Patients with chronic diseases were investigated in 25 studies^{28,35,37,38,41,42,46,49,66,68,69,71,74,77,79,82,85}, whereas a smaller number of studies (n = 6) looked at oncology patients^47,61,75,80. Only two studies^29,88 investigated low HL in refugees. Among the 79 studies reporting it, the age of the sample varied, from 45 years or less in 17 studies^{30,43,49,56,57,59,64,69,78}; between 46 and 64 years in 47 studies^{26,29,30,34,36‐38,40,42‐45,49‐51,54‐57,59,63,67,70,71,73,75,79‐82,84,85}; and over 64 years in 15 studies^{31,33,35,39,41,46,62,66,68‐70,72,74,77,78}. Lastly, the vast majority of studies (n = 81) were rated as high quality^{10,27,29‐35,38‐40,42,43,45‐52,55‐58,62‐81,83‐86}. Eighteen studies were judged as being of fair or poor quality^{26,28,36,37,41,44,53,54,59‐61,82}; their main deficits were a lack of justification for the sample size and a lack of comparability between participants and non-participants (data not shown).

Overall, the pooled prevalence of low literacy varied, depending on the assessment method used. Among tools with self-reported comprehension items, low HL was present in 42% (95% CI: 36–48%; df = 37, Q = 3451.2, I² = 99.5%); reading or numeracy comprehension items was 42% (95% CI: 33–53%; df = 28, Q = 2471.8, I² = 99.4%); word recognition items was 27% (95% CI: 18–38%; df = 22, Q = 975.5, I² = 98.3%); and mixed methods provided a pooled estimate of 48% (95% CI: 41–55%; df = 8, Q = 328.0, I² = 97.7%) (Table 2 and Supplementary Figures 1, 2, 3, and 4).

There was variation in the number of studies in each country as well as the assessment method used. Austria¹⁰, Belgium⁴⁵, Bulgaria¹⁰, Croatia⁶⁸, Czech Republic⁸⁷, Finland⁵⁰, Hungary⁸⁶, Lithuania⁴³, and Poland¹⁰ had data available only from one study each. Low HL was common: Austria, 56% (95% CI: 53–53-59%60%); Belgium, 41% (95% CI: 40–42%); Bulgaria, 62% (95% CI: 59–65%); Croatia, 58% (95% CI: 48–67%); Czech Republic, 44% (95% CI: 35–53%); Finland, 36% (95% CI: 31–42%), Hungary, 41% (95% CI: 35–46%); Lithuania, 33% (95% CI: 30–36%); and Poland, 45% (95% CI: 41–48%). Other EU countries had more than one study, often with different assessment tools. Denmark^32,35,51,52 low HL was around 44%, in both the assessment methods used (95% CI: 32–58%, df = 1, Q = 12.5, I² = 92.0%, and 95% CI: 35–54%, df = 1, Q = 0.1, I² = 0.0%, respectively) (Table 2). France^28,82 low HL was 51% (95% CI: 34–67%). Germany^{10,72,77‐79,83,84} ranged from 44% (95% CI: 38–51%, df = 11, Q = 681.4, I² = 98.9%) to 46% (95% CI: 43–49%); Greece^10,71 was 45% (95% CI: 42–48%) to 54% (95% CI: 45–63%). In Ireland^{10,30,37,38,41,42,49}, the pooled estimates varied between 19% (95% CI: 17–22%, df = 4, Q = 7.6, I² = 46.8%), 40% (95% CI: 37–43%), 41% (95% CI: 21–65%; df = 3, Q = 194.9, I² = 98.4%), and 65% (95% CI: 46–81%). Italy low HL pooled estimates were by self-reported comprehension items^61,65,66 42% (95% CI: 33–51%; df = 2, Q = 12.4, I² = 84.5%); by reading or numeracy comprehension items^61,66,67, 38% (95% CI: 35–41%; df = 2, Q = 1.5, I² = 0.0%); by word recognition items^64,65, 72% (95% CI: 32–93%; df = 2, Q = 70.8, I² = 97.9%); and by mixed method⁵⁸, 54% (95% CI: 51–57%). As for Portugal^{54,55,57,59,70}, the low HL prevalence estimates varied between 21% (95% CI: 8–46%, df = 5, Q = 86.1, I² = 96.5%), 29% (95% CI: 6–73%, df = 4, Q = 139.1, I² = 98.2%), and 50% (95% CI: 48–52%). In Spain^{10,60,63,69,89}, the prevalence estimates of low HL were, in increasing order, 33% (95% CI: 6–80%, df = 1, Q = 55.8, I² = 98.2%), 43% (95% CI: 34–52%), 58% (95% CI: 55–61%), and 71% (95% CI: 47–87%). In Sweden^36,53, low HL ranged from 21% (95% CI: 14–30%) to 39% (95% CI: 36–43%). As for The Netherlands^{10,73‐76,80,81,85}, the highest pooled prevalence was 68% (95% CI: 53–79%; df = 5, Q = 101.0, I² = 98.5%), followed by 29% (95% CI: 26–32%), 19% (95% CI: 16–23%; df = 2, Q = 2.5, I² = 0.0%), and 14% (95% CI: 12–15%; df = 1, Q = 0.3, I² = 0.0%). The UK^{31,33,39,40,44,46‐48,90} had relatively low pooled estimates, varying between 16% (95% CI: 12–20%, df = 3, Q = 48.5, I² = 92.4%), 21% (95% CI: 9–43%, df = 3, Q = 50.8, I² = 97.2%), and 28% (95% CI: 17–43%, df = 4, Q = 337.2, I² = 99.4%). Lastly, the refugees’ population was investigated in two studies^29,88 with a low HL pooled estimate of 65% (95% CI: 62–69%, df = 1, Q = 2.6, I² = 60.7%).

Meta-regression analysis found that the geographic region, assessment method, target population, and study quality impacted the results (Table 3). Western, Southern, and Eastern countries had higher rates of low HL compared to Northern EU countries. Lower quality studies found higher rates of low HL. There was no difference in literacy by the population assessed (chronic disease, oncology) or the age of the patient. Refugees had the lowest HL.